Historically, every single attempt to automate human activity has been met with stiff resistance.
In the late 19th and early 20th century, when tall buildings started coming up, lifts or elevators were operated through pulleys and levers by lift operators. This was obviously an inefficient and risky approach. Humans make mistakes. But when technology improved, and buildings tried to replace manually operated lifts with what we now know as lifts, i.e., automated rooms that go up and down a building – people simply refused to trust it. There was a long and sustained campaign by the companies that made lifts to make people aware of it. They also introduced features that give real or imagined control to people. These included a red button for emergency stop and a telephone inside. The up/down buttons on the outside that you use to call the lift to where you are though, is not always a real request. In some advanced buildings these days, that is a dummy button to give illusory control. The elevator algorithm schedules stops in the most optimal manner. Clearly, we have come quite a long way from pulleys and levers operated by actual people.
Another related if not identical example of such automation is movie recommendation engines and other product recommending systems. For example, the movie streaming site, Netflix, recommends movies to you based on what your taste is. The company even ran a famous contest for Data Scientists and Engineers to pick the best algorithm for such recommendation engines. The outcome is that we now enjoy a much better experience when we watch movies on such streaming services. The same applies to when you buy things on an e-commerce site. What’s shown to you is what the algorithm thinks you would like to buy.
However, we at FundsIndia are neither in the movie streaming business, nor in the business of selling of mobiles/books online. We are a financial services company that provides advisory services to our investors. The question to a Data Scientist in such a set-up is: are the other industry examples and the general trend towards automation relevant to what we do? Research that gives an in-depth analysis of every fund is what investors need. That is indisputable. But is there a case for implementing an algorithmic recommendation service as an additional option to our investors? If there is, how will it work?
The simple idea behind a recommendation engine is that people tend to do and like what people like them do and like. For instance, if we both like comedy movies a lot and we also like the same kind of comedy movies, you having liked a movie that I have not watched yet is a recommendation for me. We do this without the aid of an algorithm in real life. It’s called word of mouth. The computer though has the benefit of matching millions of such profiles, and then arriving at its recommendations. It’s simply unlikely that you’ll know as many like-minded people as there are in a large database like Netflix. Can we extend this idea and, say, give you recommendations of funds? Is it even the case that people who have similar investment patterns will seek similar investment vehicles in the future as well?
For instance, a person who has a balanced portfolio will not necessarily be looking to invest in more funds just because another person with a similar profile has invested in other funds. Or, portfolio balance is a limiting factor in building recommendation engines for investors. But what if we can quantify the degree of your portfolio balance and then, recommend only those kinds of funds that are missing from it? What if those recommendations are socially learned as we just saw in other examples? Will that work? Or will it face even stiffer resistance than lifts did in early 20th century? These are questions that have occupied my mind for a while and I don’t know the answers. If you do, please let us know how we can build that lift!